Abstract
The ability to use 16S rRNA gene sequence data to train machine learning classification models offers the opportunity to diagnose patients based on the composition of their microbiome. In some applications, the taxonomic resolution that provides the best models may require the use of de novo OTUs whose composition changes when new data are added. We previously developed a new reference-based approach, OptiFit, that fits new sequence data to existing de novo OTUs without changing the composition of the original OTUs. While OptiFit produces OTUs that are as high quality as de novo OTUs, it is unclear whether this method for fitting new sequence data into existing OTUs will impact the performance of classification models relative to models trained and tested only using de novo OTUs. We used OptiFit to cluster sequences into existing OTUs and evaluated model performance in classifying a dataset containing samples from patients with and without colonic screen relevant neoplasia (SRN). We compared the performance of this model to standard methods including de novo and database-reference-based clustering. We found that using OptiFit performed as well or better in classifying SRNs. OptiFit can streamline the process of classifying new samples by avoiding the need to retrain models using reclustered sequences.
Introduction
To understand the human gut microbiome, researchers typically group similar 16S rRNA gene sequences into Operational Taxonomic Units (OTUs)—functional "bins" representing microbial groups. The gold standard for creating these bins is de novo clustering, which groups sequences based on their similarity within a specific dataset.
However, de novo clustering is inherently unstable. When new patient samples are added to a study, the entire dataset must be re-clustered to maintain consistency. This process often redefines the OTU boundaries, meaning that a machine learning (ML) model trained on the original OTUs becomes obsolete and must be retrained. This lack of stability is a major barrier to deploying microbiome-based diagnostics in clinical settings, where a single, validated model is required. This paper evaluates OptiFit, a tool designed to fit new sequences into existing clusters, as a solution to this stability problem.
Methods
The study compared two primary workflows to determine if fitting new data into existing OTUs impacts diagnostic accuracy:
- The Standard (OptiClust): A de novo approach where all data (training and testing) is clustered together. The ML model is trained and tested on these shared OTU definitions.
- The Streamlined (OptiFit): Only the training data is clustered de novo. When evaluating the model, new "test" sequences are "fitted" into the existing training OTUs using the OptiFit algorithm.
Researchers utilized a Random Forest machine learning model (via the mikropml R package) to classify samples. Clustering quality was assessed using the Matthews Correlation Coefficient (MCC), and model performance was evaluated using the Area Under the Receiver Operating Characteristic curve (AUROC).
Dataset
The study utilized a publicly available dataset of 16S rRNA gene sequences from stool samples of 490 subjects, consisting of 261 healthy controls and 229 subjects with Screen Relevant Neoplasia (SRN). SRN includes both advanced adenomas and colorectal carcinomas. The data was randomly split into 80% training and 20% testing sets, with the process repeated 100 times to ensure robust results.
Key Results
Cluster Quality and Sequence Retention
A common concern with reference-based fitting is that novel sequences in new samples might be discarded if they do not match the reference. The authors found that 99.9% of sequences from the test sets successfully mapped to the training OTUs. Furthermore, the quality of the OTUs remained nearly identical between methods; the de novo approach achieved an average MCC of 0.884, while the OptiFit approach averaged 0.879.
Diagnostic Performance
There was no statistically significant difference in the model's ability to detect SRN between the two methods. During cross-validation, the AUROC was equivalent (p = 0.13), and performance on the held-out test data remained equivalent (p = 0.38). These results indicate that OptiFit maintains the high resolution of de novo clustering without sacrificing predictive power.
Discussion and Clinical Implications
The primary advantage of OptiFit is its stability. By using a study’s own training data as the reference—rather than a generic public database—OptiFit allows for a "plug-and-play" diagnostic. A model trained once can be used indefinitely to classify new samples without the computational overhead of re-clustering historical data.
While these results are promising for colorectal cancer screening, future work is needed to determine the minimum training set size required to build a sufficiently robust reference for broader populations. Additionally, the workflow should be tested across other microbiome-associated conditions, such as inflammatory bowel disease, to ensure the fit-to-reference approach remains effective across different microbial signatures. This framework provides a critical step toward the development of regulated, non-invasive clinical microbiome tests.
Summary updated